T.W. Robbins
M. James
AM. Owen*
B.J. Sahakian*
L. Melnnes
P. Rabbitt!
* Department of Experimental Psychology,
University of Cambridg
» Department of Psychology. National
Hospital for Neurology and Neurosurgery,
London,
« Northeast Age Research, Department of
Psychology, The University of Newcastle
upon Tyne, Neweastle,
Age and Cognitive Performance Research
Centre, University of Manchester, UK
KeyWords
Memory
Attention
Cognition
Cognitive test battery
MMSE
Aging
Alzheimer’s disease
Parkinson's disease
Factor analysis,
CANTAB
Introduction
—~
Review
Dementia 1994;5:266-281
Cambridge Neuropsychological
Test Automated Battery (CANTAB):
A Factor Analytic Study of a Large
Sample of Normal Elderly
Volunteers
Abstract
‘The CANTAB battery was administered to a large group (n = 787) of elderly
volunteers in the age range from 55 to 80 years. This battery, which is based on
tests used to identify the neural substrates of learning and memory in non-
human primates, has now been extensively used in the assessment of various
forms of dementia and also validated on patients with neurosurgical lesions of
the frontal and temporal lobes. The tests employed were pattern and spatial
recognition, simultaneous and delayed matching to sample, learning of visuo-
spatial paired associates, a matching to sample, reaction time task and a test of
spatial working memory. The sample was banded into different 1Q bands
based on performance on 5 standard tests of intelligence. The MMSE was also
administered to exclude cases of possible dementia (n = 16) in the normal
sample, In general, performance declined with age and 1Q, but these factors
did not interact, A factor analysis (with varimax rotation) identified 4 factors
with eigenvalues greater than 1, which accounted for over 60% of the variance.
Factor 1 was equated with general leaming and memory ability and loaded
significantly with the Intelligence scores; factor 2 was related to speed of
responding and loaded most heavily with Age. Comparisons were also made of
performance on CANTAB of those subjects with dementing scores on the
MMSE and the lowest 5th percentile of the population sample. The results are
discussed in terms of the utility of the CANTAB battery for the assessment of
dementia and of the implications for theories of changes in cognitive function
during normal aging.
the frail and confused elderly. This allows the many other
advantages of computerised testing to be realised: com-
In recent years, methods of testing cognitive function
in elderly and dementing patients have been augmented
by batteries of computerised tests [1-3]. New technology,
such as the use of touch-sensitive screens rather than key-
boards, have been used to make these accessible, even to
Asset
Sanus 1, (94
puterised tests can immediately check the accuracy of
each response made and so provide patients with instant
feedback; they can continuously check to ensure that task
demands are correctly understood and can continuously
adjust levels of difficulty to challenge, but not to embar-
FW Ratios
‘Deparment af ExperentalPyehlosy
Unieiy of Cambridge
Donen St
Cambridge, CB2 3EB (UK)
01900 Kamer AG. Ba
1o13-ra20s.20055-0266
staoo rass them; from any task it is possible to derive, record
and automatically analyse several different performance
indices, rather than merely one. Direct feedback and
appropriately graduated levels of challenge have been
shown to increase interest in tasks and so promote moti-
vation to do well. This mitigates the perennial problem in
assessing elderly patients that not only the limits to their
abilities but also their levels of motivation may be
reduced by age or pathology. Computerised tests offer
neutral settings and highly consistent modes of presenta-
tion that reduce the large experimenter effects that often
reduce the validity of assessments. All of these factors
obviously improve the reliability and comparability of
results across studies,
Cambridge Automated Neuropsychological Test Bat-
tery (CANTAB) was designed to exploit these advantages
[4-7]. The original theoretical impetus was to adapt para-
digms developed for testing animal models of dementia in
order to relate the findings to man. To do this it was neces-
sary to find ways of varying task demands appropriately
toa very wide spectrum of ability. However, as the battery
was developed, tests based on methodological advances in
neuropsychological assessments of human patients were
also added: for example the Visual Memory section of the
battery originally consisted of tests of pattern and spatial
serial recognition memory, and a delayed matching to
sample (DMTS) procedure analogous to those used by
Gaffan [8] and Mishkin [9] and their colleagues to deter-
mine the neural basis of visual recognition memory in
infrahuman primates [see also 10]. Based on the delayed-
(non)-matching-to-sample paradigm, Mishkin [9] has pro-
posed a neural model of visual recognition memory that
includes as important elements, structures of the temporal
lobe, including the hippocampus. To these tests have been
added a more complex visuospatial paired associated task
in which subjects have to learn the spatial locations of dif-
ferent patterns, analogous to conditional associative
learning paradigms used by Petrides [11] to test both
monkeys and human patients with temporal, but also
frontal lobe damage. Finally, in this battery we have
employed a test based somewhat on the 8 arm radial maze
test to assess ‘working memory’ in rodents [12], but more
particularly in non-human primates receiving circum-
scribed lesions of the prefrontal cortex [13]. We have
already shown that performance on this test is impaired in
patients with neurosurgical lesions of the frontal cortex
im.
‘Because many of these tests make demands on percep-
tual information processing, on attentional resources and
on central executive function, it is important to design
memory tests which control for these factors and so can
reveal independent, residual changes in memory func-
tion. For example, in the DMTS test in the CANTAB bat-
tery the contribution of attentional changes to task perfor-
mance is assessed by including a simultaneous matching
to sample task. Similarly, possible contributions of
changes in information processing load have been taken,
into consideration by measuring reaction times to match
remembered stimuli to presented samples and by system-
atically varying task difficulty by increasing the sizes of
the sets of distractors among which target stimuli are
embedded. Executive function, which determines the
strategies that subjects may adopt in particular memory
tasks, is assessed independently of these tasks by inclusion
of a spatial working memory task in which subjects have
to develop and use their own strategies to search through a
set of spatial locations.
‘The CANTAB batteries have been extensively used to
assess cognitive performance in a number of neuropsy-
chiatric disorders including dementia of the Alzheimer [5,
14-16}, Lewy body [17, 18] and frontal [19] types, Korsa-
kofi’s syndrome [20], depression in the elderly {21 basal
ganglia disorders such as Parkinson's disease [5, 6, 20,
24], progressive supranuclear palsy [25] and multiple sys-
tem atrophy [26]. The validity of the tests in distinguish-
ing between neurosurgical populations with localised ex-
cisions of the frontal or temporal lobes has also been dem-
onstrated [7, 27, 28]
In all these studies it has been possible to compare
patients individually to appropriately matched normal
controls. This has been satisfactory for the comparisons
intended but has not allowed us to determine the general
background effects of individual differences in factors
such as age, gender and intelligence independently of the
pathology examined. Itis of obvious practical importance
to obtain baseline data on large groups of normal individ-
uals in a wide range of older age groups against which the
performance of various groups of patients can be as-
sessed.
The lack of such data is also a restriction to develop-
ment of theories of individual differences. A lively re-
search topic in current cognitive psychology and gerontol-
ogy is the extent to which the cognitive concomitants of
individual differences in age, or in general intelligence can
be accounted for in terms of changes or individual differ-
ences in the efficiency of particular brain structures (29,
30). For these reasons we administered the CANTAB bat-
tery toa group of 787 normal elderly volunteers who had
all been repeatedly assessed, over a period of 7 years, on
two other, different batteries of cognitive tasks. These
267 Table 1. Characteristics of population by age
Agegroup a Mean age Mean NART Mean INTEL
35-59 79 S772) 118.6(6.6) 40.28
Male 10 57.2(09) 118.4(8.6) +047
Female 69 $7.8(1.3) 118.6(6.4) +0.25
60-64 123 62.3(1.4) 118.0(6.5) +019
Male 20 © 61.911.) 119.6(6.0) +047
Female 103 624(L4) 117.7166) +014
65-69 222 67.3(14) 117.9(6.0) +004
Male 56 G7.2(1.5) 119.3(6.0) +033
Female 165 673(L4)117.5(59) -0.05
70-74 219 72.0(1.5) 117.8 (60) 0.04
Male 65 23(14) 18.0(6.1) +018
Fomale 154 TLB(LS) 117.7160) 0.13,
15-19 128 76.6(1.4) 116.2163) -022
Male 40 76.4(1.5) 117.7(54) +017
Female 88 76.7(4) 115.6(6.6) 0.39
NART = National Adult Reading Test; INTEL = mean weighted
2z score ofthe intelligence tests administered (minus scores, low; posi-
tive scores, high IQ), standard deviations in parentheses.
Table 2. Characteristics of population by IQ band
1QBand =n Meanage Mean NART Mean INTEL
Qi 24 69.257) 111.06.)
Male 38 1103.6.)
Female 186 126.)
1Q2 235 69.1(59) 1175.2)
Male 60 715(54) 1170140)
Female 17868388) 117.7(3.70)
193 312 67.3(59) 122.527)
Male 93 68.9(56) 122.827)
Female 219 66.6(60) 122.527)
NART = National Adult Reading Test; INTEL = mean weighted
2z score of the intelligence tests administered (minus scores, low; posi
tive scores, high IQ); standard deviations in parentheses.
additional data allowed us to extend our analyses to evalu-
ate variations in CANTAB scores with individual differ-
ences in age between 55 and 80, and in performance on
five different measures of general intelligence, in the con-
text of variations in scores on a variety of other measures
of cognitive ability.
268
Rabbitt
Robbins/James/Owen/Sahakian/MeInnes!
Method
‘Subjects. Subjects were 787 normal healthy residents of Newcas-
Hle-upon-Tyne aged from 55 to 80 years. These had volunteered to
take part ina large scale longitudinal study of the cognitive effects of
normal ageing funded by the Medical Research Council and the Eco-
nomic and Research Councils and carried out at the University of
‘Newcastle-upon-Tyne. Over a Teyear period prior to this study all,
baad been repeatedly assessed on three different 1Q tests: National
‘Adult Reading Test, NART [31], Mill Hill Vocabulary Test, ‘A’ and
“B' [32], and AHI and AH42 [33]. This allowed the subject popula-
tion to be divided orthogonally, for purposes of data analysis, into $
different age groups (table 1) and 3 different IQ test score groups (ta-
Dle2) in terms of their attainment on the several different 1Q mea
sues, For the purpose of obtaining a single IQ measure, we trans-
formed all 1Q test scores to zscores and the resulting measure,
“INTEL i the average of these scores and is summarized in table 2.
‘The highest 1Q band group is called 1Q3, the middle group, 1Q2 and
the lowest group, IQ
To detect and independently study any individuals who might be
at risk of dementia all volunteers were also screened on the Mini
Metal State Examination (MMSE) [34]. Al volunteers were then
tested on all the subtests of the CANTAB battery as deseribed below.
The main analysis of results was carried out on the 771 individuals
‘who attained MMSE scores greater than 23, Dats for the 16 individu-
als with MMSE scores less than 24 are presented separately.
Equipment and Procedure
The main tests were taken from the CANTAB. These were 2
series of computerised tasks run on @ Acom BBC Master 128 micro-
computer with a resolution Microvitec 12-inch VDU and a Microvi-
tec Touchtec 501 touch-sensitive sereen, On one of the tests (match-
ing-to-sample - visual search ~ task), subjects were trained to depress
a large key-pad interfaced into the serial port of the microcomputer
With their hand for the purpose of measuring reaction times. Subjects
sat at a comfortable height approximately 0.5 m from the monitor. It
‘was explained that they would have to respond to stimuli by touching
the screen. They were introduced to the apparatus by way of a task
‘which screened them for possible difficulties in motor ability that
right affect their performance. In this ‘motor sereening task’ they
were asked to respond to a series of flashing crosses on the screen by
placing the index finger of their preferred hand on the centre point of
each cross as soon as possible after it appeared. Once each eross had
bbeen accurately touched the next cross appeared after a 6-second
delay. Following a short demonstration by the experimenter, in
hich three consecutive crosses were touched, subjects were present-
ed with a series of ten crosses (0 touch. All volunteers passed this
introductory sereening task and were then given the following tests in
the order described.
Pattern Recognition. This test was presented in two phases. Sub-
[jects were frst shown a series of 12 coloured patterns (set 1) which
‘appeared one at a time, for 35, inside a white box located in the
centre ofthe sereen, After each pattern the screen was cleared and the
next appeared. In the second (recognition) phase, 12 pairs of
coloured patterns appeared successively, and one at a time, One
‘member of each pair was a pattern that had appeared in set one (tar-
‘get pattern) and the other was a novel pattern. During the recognition
phase the target patterns appeared in the reverse order of their origi-
nal presentation and the distractor patterns differed in form bat not
in colour from members of the target set. Subjects responded to each
CANTAB:
Sample
actor Analysis ofa Large pair by touching the pattern they recognised as having been a mem-
ber of the target set. The programme signalled satisfactory registra-
tion of each response by sounding a tone. Visual feedback of accuracy
of each response was automatically provided by appearance of @
green tick or red cross. This cycle was then repeated with 12 new
patterns (set 2) and the subject's total score (maximum possible = 24)
was expressed as the percentage of correct to total responses made.
‘Spatial Recognition. In the initial (presentation) phase a series of
S unfilled L-inch white squares appeared for 3s, one at atime, cach at
«different spatial location on the screen. In the second (recognition)
phase, two squares appeared simultaneously. One, the target, oceu-
pied a location used during the presentation phase. The other, the
distractor, appeared ina previously unused location, Thesubject had
to recognise and tap the target. During the recognition phase the tar-
gels appeared in the reverse order in which they had appeared during
presentation, The programme signalled that it had registered each
response by a tone, and provided visual feedback of accuracy by
jreen ticks and red crosses. This procedure was then repeated 3 more
times, using new target and distractor locations on each occasion.
‘The subject's score was expressed as the pereentage correct of total
responses made (ie. 20)
‘Simultaneous and Delayed Matching to Sample. Atthe beginning
ofeach tril, a complex abstract pattern consisting of four quadrants,
each differing in colour and form, appeared in the centre of the
screen for a period of 4.5, Subjects were told to study the pattern,
since they would later be required to identify it among three ‘distrac-
tor’ patterns. In the simultaneous condition, 4 choice patterns then
appeared, located under the sample pattern. The subject was re-
quired 10 respond by touching the choice patter that exactly
‘matched the sample pattern in both colour and shape. On each dis-
play only one of the choice patterns matched the sample. One of the
other choice patterns was a novel distractor, differing from the sam-
plein both colour and shape. One had the same coloursas the sample
but the same shape as the novel distractor while the other had the
same shape as the sample but the same colours as the novel distrac-
‘ot, To discourage mnemonic strategies based on remembering the
colour and shape of a single quadrant, each of the 4 choice patterns
had one, randomly chosen quadrant in common [fig. 1 of ref. 24).
Registration of each response was signalled by an auditory tone and
areen ticks and red crosses provided visual feedback of accuracy,
After an incorrect response subjects had to continue to choose until
they touched the correct target stimulus. The interval between the
appearance of the 4 choice patterns and the subject's response (reac-
tion time), whether correct or incorrect, was timed in ceatiseconds,
‘The delay condition was identical to the simultaneous condition
‘except that the sample pattem disappeared from the sereen imme
diately after the intial 4.5-second study period. A 0-, 4 or I2-second
‘delay then followed before the display of 4 choice patterns appeared
and subjects were required to make their selection. Following three
practice trials (one each of simultancous, 0 and 125), there were a
total of 10 test trials on each of the four simultaneous and delay con-
ditions presented sequentially, and in a pseudo random order (total
test trials = 40).
In each of the simultaneous and delay conditions, the score was
the number of correct recognitions made on the frst choice in each of
the simultaneous and delay conditions. Errors in each ofthe four test
conditions were also analysed to determine which of the three types
of distractors had been incorrectly chosen (shape, colour oF unce-
lated), Finally in each condition, mean RTs were computed for those
trials in which the first choice had been correct.
Visuospatial Paired Associates Learning. tn this test, subjects
were required to remember up to 8 pattern-location associations. I
tially, 6 solid white boxes were presented around the screen [fig, 1 of
ref, 24] and subjects were told that these would ‘open up’ in turn,
showing them what was inside, Their task was to look for coloured
patterns in the boxes and to remember which pattern belonged in
‘which box. Each box opened for 3 sand closed again before the next
box opened, one by one in a randomized sequence. On the first tral
only one of the boxes contained a coloured pattern, Immediately
after the las box had opened this pattern was presented in the centre
of the screen and the subject was required to respond by touching the
box in which it had appeared. Feedback was not provided after each
response although, i all choices had been correct, the words ‘all cor-
treet’ appeared in the centre ofthe screen and the subject proceeded to
the nent trial, Ifthe choice had been incorrect, the boxes were succes-
sively reopened and shut for 2 seach (remvinding phase). The subject
1was then given a second attempt to correctly locate the pattern. On
cach trial, subjeets were allowed up to nine veminding phases, mak-
ing ten attempts in all. I'they failed all ofthese, the series of tests was
stopped at this point and the conditions experienced were scored as
described below, As soon as any of these nine possible further
attempts had been successful the test moved forward to the next
stage, After the initial stage with one pattern, there was one further
stage with a single patter, then two stages with two patterns each,
too stages with three patterns each and then one stage with six pat-
tems (i. one in every box). Finally, two extra boxes were added to
the array on the sereen and the subject was required to correctly
locate a total of eight patterns. Subjects automatically moved from
fone stage to the next by correctly locating all of the pattems, either
after the intial presentation phase or after one of the nine reminding
phases.
Performance was scored using three indices: (1) Trials; the total
‘number of presentations required (maximum score = 10 presenta-
tions per trial) to correctly locate all of the patterns summed across
each of the eight stages. Subjects were assigned the maximum score
of 10 for stages not attempted due to failure at an earlier stage.
(@) Errors: the total number of errors (incorrect placements) summed
across the eight stages. (3) The Fist trial corvet memory score was
the total number of patterns correctly located alter the first presenta
tion, summed across the eight stages (range = 0-26),
Spatial Working Memory. Subjects were required to ‘search
through’ a number of boxes presented on the screen in order to collect
“blue tokens’ hidden inside (fig. 1 of ret. 23]. On any one display pre-
sentation a single token would be ‘hidden’ inside oneof the boxes and
subjects were to search, by touching and “opening’ boxes in turn, until
they found it. At this point the next token would be hidden, Once @
blue token had been found within a particular box, that box would
never again be used to hide a new token. Errors were scored accord
ing to the number of occasions on which a subject returned to open a
box in which they had already found a blue counter during previous
search, The performance index analysed was the total number of
those errors summed over twelve trials on this task (4 each with 4, 6
and 8 boxes).
Matching to Sample (Visual Search for Designated Targets). &
central red box surrounded by cight white boxes appeared on the
screen, To initiate each trial, the subject was required to hold dawn a
key pad which required minimal pressure. Once the subject depressed
the key pad the boxes opened to reveal the central target stimulus
surrounded by the choice stimuli among which an identical match
hhad to be located. To begin with the subject was shown the arrange-
269 ‘ment and told to release the key pad as fast as possible and touch the
{identical stimulus inthe peripheral boxes to the central sample stimu-
lus. Feedback was provided on the screen for correct and incorrect
responses. On an equal proportion of trials there was 1,2, 4or 8 differ-
‘ent patterns to choose Irom. There were 4 examples of increasing set
size, before beginning the 48 test trials. The test stimuli resembled
Pattern recognition
1007 8
Spatial recognition
those used in the delayed matching to sample test. Each of the test
stimuli was made up of 4 quadrants which varied according to colour
land pattern, but the same 4 colours were used for a given set of
choices. For set sizes greater than 1, half ofthe stimuli were derived
from the target stimulus and half from the distractor stimulus, by
-varying the relative positions of the quadrants (see fig. 2 of ref. 6}
cso)
i coca a)
BD 65.69 (222)
7074218)
D 75.79 128)
Bo
8 os
e
|
| 6
| «
Test
Pattern recognition ‘Spatial recognition
wy b
Fig. Pattern and spatial recognition
Ho 22) memory accuracy performance as funetion
Ol rozass of age (a) and intelligence score (). Values
D ns are given as means + SEM. The lines beneath
ew the histograms summarise the results ofthe
z post hoe tess following analysis of variane.
5 Groups with overlapping lines do not differ
3 significantly from one another, those groups
2 not sharing an overlapping line are signifi
5 cantly different.
g 04 Fig.2. Simultaneous and delayed match-
é ing to sample performance. Accuracy and
latency measures as a function of age (a, b);
: asa function of intelligence score, @). Both
the accuracy scores and latency scores were
FA transformed prior to statistical analysis and
these variables are pletied on the appro-
riate scale. Other conventions as in fig-
ure
270
Rabbitt
Robbins/James/Owen/Sahakian/MeInnes!
‘CANTAB: Factor Analysis ofa Large
‘Sample Proportion correct (aresin transform)
Proportion correct (arcsin transform)
—— sss009
a siya
si
er
)
nae
seen
fone |
mt
100
orf ¥ so
g
ous S
Bom)
B ase f
on : i
aa Ban
om im
ou 3 te a : te
sim DELAY SIM DELAY
om | —— ee |
=e (ete, | oe 0225) |
aise | Bib
x
soa
& 436 f
om i
Pua {
ae
on
oa
om
a : 7 Tes ri i 7 ee
Delay Delay
271 —
Table 3. Two-way ANOVA: summary
Age Intelligence A x 1
banding banding Fagoe
Fan Frases
Pattern recognition 3.96 1.70
Spatial recognition 5.95 037
‘Simoltaneous MTS 3.90 07
Paired associates
Total rials 15.08 12.86 109
Total errors 14.40 10.67 08s
Memory score 1368 12.40 133
Spatial working memory 1344 2470 ost
Table 4. Three-way analyses of variance for DMTS: summary
Accuracy Latency
F a F dt
Age 13.35 792 4,717
Intelligence 35.22 146 2,717
Delay 200 21512275892, 1434
Axt 123 8,756 063 8,717
AxD 186 8.1512 087 8.1434
IxD Lag 41512 183 4.1434
AxIxD 1.00 16,1512 0.73 16,1434
Results
‘The profile of performance as a function of age and 1Q
group is illustrated for each test with figures summarising
differences between age and intelligence groups and ta-
bles 3 and 4 summarising the main effects and interac-
tions found in analyses of variance (ANOVAS).
Pattern and Spatial Recognition
Accuracy of performance declined with age on both the
pattern and spatial recognition tasks as shown in figure La
although at slightly different rates as evidenced from the
ost hoc comparisons made between each age group,
depicted in the figure, Scores declined smoothly and sys-
tematically from high to low IQ groups (fig. 1b).
Separate analyses of the effects of Gender (covarying
for both age and intelligence score) found no difference
between men and women for pattern recognition (F),209 =
1.81, p> 0.05), but a small significant male advantage for
spatial recognition (F; 299 = 4.18, p < 0.05) (mean scores;
males 79.55%: females 77.65%).
Simultaneous and Delayed Matching to Sample
Accuracy on the simultaneous matching condition de-
clined significantly with age (fig. 2a). However it isimpor-
tant to stress that overall levels of performance were exce!-
lent, and that even the oldest age group attained 93%
accuracy. Latency scores show a markedly different pat-
tem (Fig. 2b). While the youngest subjects were no more
accurate than the 60- to 64-year-olds, they were signifi
cantly faster. Moreover, although the two oldest age
groups (70-74 and 75-79) were equivalently fast, the 75-
to 79-year-olds were significantly less accurate. These
contrasts imply differences in speed/error trade-off func~
tions with age. There was a significant effect of intelli-
gence band for both accuracy and latency (table 3,
fig. 2a,b). For both measures, the high IQ group (1Q3)
performed significantly better than either the middle
(1Q2) or the low (IQ1) groups, who did not differ. An anal-
ysis of gender effects, covarying for both age and IQ test
score, found that female subjects were slightly, but sign
cantly faster at simultaneous matching to sample (F399 =
4.18, p < 0.05).
In the delayed matching to sample condition, as ex-
pected, performance declined as a function of delay dura-
tion, Although overall accuracy did decline with increas
ing age, there was no significant difference in the rate of
decline in accuracy with increasing delay intervals as a
function of age (table 4). However the delay X age group
interaction approached significance (p = 0.064) and so it,
was thought appropriate to consider the simple main
effects of age group as function of delay. Inspection of fig-
ure 2a shows that performance of the two youngest age
groups is indistinguishable except at the longest (12)
delay. In contrast, for the two oldest age bands, a signifi-
cant difference was found at short delays, but not at the
longest delay. Counts of the different kinds of errors made
found that in 74% of cases the distractor was of the same
colour but different shape, in 21.5% of cases the distractor
was of the same shape but different colour, and in 4.5% of
cases the random distractor was chosen in error. Overall,
it seems that there is some evidence to support the
hypothesis that short-term forgetting rate increases with
age and that, possibly because of encoding strategies that
the subjects used, some stimulus aspects are less well
encoded, or more rapidly forgotten than others.
Latencies showed a similar pattern of effects. Latency
increased as a function of delay interval indicating that,
like accuracy, itis a good index of efficiency of short-term,
retention. However, unlike the findings for accuracy, the
age x delay interaction fell far short of significance. Fig-
ure 2b illustrates this, showing that all age groups showed
m
Pabbitt
Robbins/James/ Owen/Sahakian/Melnnes!
CANTAB: Factor Analysis of a Large
Sample Trials
Errors Memory score
5559(79)
6054 (129)
65:69 (222)
7074 (819)
75:78(126)
oposm
2»
0
Trials Errors Memory score
wb
2 Mose)
D raze)
DB sei)
20 —
Fig. 3. Visuospatial paired associates. £
Summariesofthe3 main measures of perfor | 10
‘mance (mean + SEM scores) plotted for age ——— __— _——
(a) and imteligence score (b) groups. Other Measure
conventions as in figure 1
in lateney with increasing delay. The sin-
gle exception is that the 60-to 64-year band are dispropor-
tionately slow at the 0-second delay. Since these subjects
made few errors at this delay interval, this perhaps repre-
sents a sacrifice of latency to increase accuracy
Differences in short-term memory performance be-
tween 1Q groups were much less clear-cut, The 1Q3 group
was significantly more accurate than either 1Q2 or QI
groups at all delay intervals, but latencies did not differ
significantly across the groups from 0 to 12 s (fig. 2c,d). ‘There were no significant two- or three-way interactions
involving Age and IQ test score group.
Visuospatial Paired Associates Learning
Even at the most difficult level of 8 items very few sub-
jects failed to reach the ‘all correct’ criterion in ten trials
‘or fewer; (14/771 = 1.8%) overall. The highest failure rate
was in the 75-79 age band (4/128 = 3.1%). Other mea-
sures of performance, including total errors (omitting
those subjects who failed to complete) or total trials to cri-
terion and the first trial memory score produced compara-
ble patterns as a function of age or intelligence group
(fig. 3a,b). In each case, this test failed to distinguish
between the two youngest groups or the two oldest groups,
although the former performed significantly better than
the latter. The middle age group (65-69) fell between
these extremes.
Spatial Working Memory
Figure 4a,b shows the effects of age and IQ group on
between search errors. All age groups, except the two
youngest, differed significantly from all others. There was
also a clear gradation of performance from 1Q3 through
1Q2 and IQI groups. There was a highly significant differ-
ence in search error scores between men and women (co-
varying for both age and intelligence score (Fy,209 = 20.21,
p< 0.001), with female subjects performing less well
(mean error scores; males, 41.71; females, 48.97.
Maiching to Sample
(Visual Search for Designated Targets)
Appreciable numbers of errors only occurred with the
largest set size of 8. The effects of age and intelligence
grouping are shown in figure 5a,b. There was no signifi-
cant interaction between Age and IQ test score. As might
be expected these effects were almost identical to those
described for the simultaneous matching to sample task,
in which similar stimuli were used.
Figure 5c,d shows latency as a function of set size.
Data were analysed separately for set size 1 and for set
sizes 2-8. For set size 1, there were significant effects of
Age (Fs.o8 = 16.90) and Intelligence (F2,693 = 11.20) on
latencies, but there was no significant interaction between
them (Fgg9s = 0.49).
‘As expected latencies increased with set size and with
increasing age (table 5). However, there was also a signii-
cant Age x Set size interaction (over sets 2-8), showing
that processing time per item was affected by increasing
age. Figure Sed suggests that the two oldest groups
showed a flattening of the function relating set size to
reaction time, especially at set size 8, This occurred in par-
allel with an increase in errors for this level of difficulty.
. m oraee
i 55-59 79) Boe BD 102235)
Be sec2 Boscia
Bess 222) r
ae 70-7218) ss
g 0 7908)
5 zl
gs ©»
2
Ds “8
«
5
al
14. Spatial working memory; between search errors, plotted Fig. 5. Visual search, matching to sample. Accuracy and latency
forage a) and intligence score groups (b) Other conventions asin variables plotted separately forage, e)andinteligence score (b,
fig |. Other conventions asin figure |
ma Robbins/TamesOwen/Sahakian/Melnnes’ CANTAB: Factor Analyisofa Large
Rabbitt
Sample Errors
Lateney see (L610)
5559079)
60s (128)
65.69 (222)
7074 219)
75:79 (128)
SeT1
28
—— 120)
2 1024205)
tos
set 1
28
12
we
os
3 A | J
0 + 00:
1 z 7 3 7 7
ser 28
ae | cr: el
Se wanes i
—— eee orem | |
a= mr. sors ||
Se reap —— we 1 |
620 so
2.08
18
2s 2st pf
tf 0 Z
4 158:
a a
I z
veo vas
Set size
Set size
275 Table 6. Three-way analyses of variance for MTS: summary
‘Accuracy Latency
F ae F af
Age 246 4,755 4,755
Intelligence 430 2,755 2,758
Set size 24080 2.1510 2.1510
Axl 148 8,755 8.775
AxS 190 8.1510 81510
Ixs 148 4.1510 4.1510
AxIxS LIL 16,1510 L3t 16,1510
Table 6. Summary of loadings for CANTAB tests on factors 1-4
following factor analysis
Factor | Factor2 Factor3 Factor4
Paired associates learning,
trials 082
Paired associates learning,
memory score on
Pattern Recognition 073
Spatial Recognition 04
DMTS (sim) accuracy on
DMTS (de!) accuracy 0.63
DMTS latency 0.76
MTS (visual search)
accuracy 0.80
MTS (visual search)
latency on
Spatial working memory
(between search) errors 088
Spatial working memory
(within search) errors om
As in the simultaneous matching to sample test (see
above), this suggests that increasing age does not merely
alter the limits of speed and accuracy that people can
attain but also, quite systematically, alters the speed-error
trade-offs (compromises between speed and accuracy)
that they choose to adopt. The effects of IQ test score abil-
ity contrast with those of age because, as the significant IQ
lest score group x Set size interaction shows, the lowest
1Q score group took relatively longer to make correct deci-
sions for set size 8. Again, there was no significant interac-
tion of age and IQ test score group.
Asin the simultaneous matching to sample task female
subjects were significantly faster than males but, in this
case, only at set size 4 (F709 = 4.79, p<0.05).
Factor Analysis (table 6)
Eleven key variables were included in the analysis. The
method employed was a Varimax rotation applied first to
the entire data set (n = 771) and subsequently to 5 random
samples, each of n = 200, in order to check the consistency
of the factor structure within the population. Eight addi-
tional analyses were undertaken; one within each of the 5
age bands and one within each of the three IQ test score
groups.
For the 5 random population samples, 4 factors were
consistently extracted with eigenvalues above 1.00 (mean
eigenvalues = 3.32, 1.65, 1.3, 1.03). These factor loadings
were averaged over the 5 random samples, and accounted
for an average of 60.52% of the variance. Table 6 gives the
mean factor loadings for the 11 key variables. Only those
variables loading above 0.50 are reported. The factor
structure was checked against that obtained when all 771
subjects were included in a single analysis. Only one dis-
crepancy was noted; for the total sample the DMTS accu-
racy scoreloaded more heavilyon Factor 4 than on Factor |
The most parsimonious interpretation of the factor
structure is that Factor | represents general learning and
memory ability, Factor 2 represents speed of responding,
Factor 3 represents executive processes including spatial
working memory and Factor 4 respresents visual percep-
tual ability.
The factor structure remained consistent when tested
separately within each age and 1Q test score group. The
main difference was that Factors 1 and 4 were readily dis-
tinguishable in all age groups except the youngest age
band (55-59). However, it should be pointed out, this
result is compromised by the relatively small n (= 79) for
this group.
‘When age and intelligence scores were included in the
analysis of all 771 subjects, these variables loaded most
heavily on Factor 2 (speed of responding) and Factor 1
(learning and memory), respectively.
Profite of Performance in the Lowest 5% of the
Population
The profile of performance was examined in those sub-
jects performing within the lowest 5% in the various cog-
nitive domains defined by the factor analysis, but exclud-
ing those subjects with scores less than 24 on the MMSE.
‘The areas of cognitive function represented by Factors 1-
4 above were defined by performance on variables with
76
Pabbitt
Robbins/James/Owen/Sahakian/Melnnes/
CANTAB: Factor Analysis ofa Large
Sample the highest loadings for each of the 4 factors. These were
as follows; paired associates learning, total trials measure
(Factor 1), latency on the DMTS (Factor 2), between
search errors for the spatial working task (Factor 3) and
errors on MTS (Factor 4). Comparisons across the 4 fac-
tors showed that most of these low IQ test scorers showed
impairment on only 1 factor and that this pattern re-
mained consistent across all 5 age bands. Thus, of 109
(14%) subjects performing below the Sth percentile on at
least 1 of the 4 factors, only 109.29) were impaired on 2
factors. This represents only 1.3% of the total population
of 771 subjects. No subject was impaired on more than 2
ofthe 4 factors.
Low MMSE Subgroup
Results of the 16 subjects with MMSE below 24, who
had been excluded from the main analysis, were analysed
separately. For Factor 1, the performance of 8 of the 16
subjects (50%) was below the 25th percentile, with only 3
of these (19%) falling below the Sth percentile; for Factor
2, the performance of 3 of the subjects fell below the 25th
percentile and none fell below the Sth percentile; for Fac-
tor 3, the performance of 13 of the subjects fell below the
25th percentile and 3 fell below the Sth percentile; for
Factor 4, the performance of 6 of the subjects fell below
the 25th percentile and 3 fell below the Sth percentile.
‘Thus of these 16 low MMSE subjects, 2 scored below
the Sth percentile on I factor, 2 scored below the Sth per-
centile on 2 factors and only 1 scored below the Sth per-
centile on 3 factors. The corresponding numbers for
scores below the 25th percentile were 5, 6 and 4 subjects
respectively. One subject performed above the 25th per-
centile on all 4 factors.
In summary, there is relatively little overlap between
those subjects scoring below the accepted cutoff for
dementia on the MMSE and those scoring in the bottom,
Sth percentile on the computerised tests.
Discussion
‘The CANTAB battery was developed to extend para-
digms developed to test animal models of the neurobio-
logical and neurochemical changes that are believed to
occur in humans who suffer from dementia of the Alz-
heimer type (DAT) and other pathologies that become
increasingly common in later life. This standardisation of
the CANTAB battery on a large, healthy population of
older people shows that scores on subtests do sensitively
differentiate between individuals of very diverse ages and
levels of general ability.
‘The most important finding is that principal compo-
nents analyses show that scores on CANTAB subtests fall
into readily discriminable factors which can plausibly be
identified with particular groups of cognitive functions
that are known to be supported by distinet neural subsys-
tems. For example, it is known that the spatial working
memory test is especially dependent upon intact frontal
lobes, whereas the pattern recognition memory test is
unaffected by frontal damage but is sensitive to lesions of
‘the temporal lobe [7, 28]. A second important point is that
the effects of individual differences in age and IQ test per-
formance do not appear as global and equivalent changes
in scores on all factors but rather as local changes in some
factors more than in others.
These findings contrast with predictions from current
models of cognitive ageing which have been developed
from principal component analyses of performance on
very small subsets of cognitive tasks. Such analyses have
generally yielded a single dominant factor which has been
functionally reified as ‘information processing rate’. This
has been interpreted as evidence that changes in a single
slobal performance characteristic of the central nervous
system, ‘mental speed’, underlie all age-related changes in
all other cognitive tasks [35, 36]. Parallel psychometric
investigations which, curiously, have never cross-refer-
enced data from cognitive gerontology, have also identi-
fied information processing rate as a functional reifica-
tion of the single factor ‘g which Spearman [37] postu-
lated as basic to all mental abilities [38-40]. This position
has recently been modified to substitute ‘working memo-
ry efficiency’ for ‘speed’ as the single, basic performance
characteristic that underlies individual differences in in-
telligence test performance [41] or in cognitive ageing
[42]. However this shift in emphasis does not seem to
indicate any relaxation of the assumption that individual
differences in cognitive ability associated with IQ test per-
formance or with age can best be described as points along,
a single continuum which can be identified with a particu-
lar, single performance characteristic of the central ner-
‘vous system.
The idea that a single simple performance characteris-
tic such as ‘information processing speed’ or ‘working
memory efficiency’ underlies performance on all or most
cognitive tasks is not well supported by the present data.
Atleast 4 distinct and well separated factors are necessary
to describe individual differences in performance on the
CANTAB battery. The Ist factor can be identified with
performance on tests of memory and learning, and ac-
27 counted for about 28% of the variance. Additional analy-
sis showed that this factor loaded on variance associated
with individual differences in current scores on IQ tests.
In contrast, the 2nd factor accounted for about 14% of the
variance and loaded for speed of response in tests of mem-
ory and selective attention. This speed factor was more
strongly associated with individual differences in age than
in IQ test scores, The final 2 factors respectively loaded
heavily on the test of spatial working memory (11% of
variance) and on the visual perceptual aspects of the tests
(8% of variance). We suspect that the loading of the 3rd
factor on spatial working memory represents an executive
function and this remains to be determined in our further
analysis of tests of frontal lobe function.
This factor analysis suggests that the CANTAB battery
does fulfit its primary aim of providing a componential
analysis of particular cognitive functions. The clearest
example of separation of cognitive abilities is the loading
of both spatial and pattern recognition with the more
complex test of visuospatial learning. The factor structure
of the battery was stable within and between all the var-
ious age and IQ test score groups, with very rare excep-
tions. This suggests that the interrelationships between
the various cognitive processes contributing to effective
performance on CANTAB, remain relatively constant
between ages 55 and 80 and across a wide individual vari-
ation in general ability, as assessed by IQ test scores.
Thus, from the point of view of interpretation of indi-
vidual differences in performance, factor analyses of the
CANTAB battery do confirm that differences between
age and IQ test score groups are associated with individu-
al differences in information processing speed and in the
ability to learn and remember new information. However
they also strongly suggest that neither of these two perfor-
mance characteristics can, on its own, provide a complete
description of the cognitive effects of ageing or of the
functional bases of intelligence.
Analyses of data from individual tests in the CANTAB,
battery reinforce the point that individual differences in
age and in IQ test score performance are better described
in terms of patterns of differential changes across discrete
performance indices rather than in terms of positions
along a single continuous measure, Some performance
indices, such as pattern recognition and reaction time,
showed regular and continuous declines in performance
with increasing age. Others, in particular spatial working
memory, paired associates learning and spatial recogni-
tion memory, showed strikingly different trajectories of
change with age. Performance on all these indices de-
clined significantly between the 55-59 and 60-64 age
bands. However, while spatial working memory contin-
ued to decline between the two oldest age groups, this was
not the case for paired associates learning. It is unlikely
that these differences in patterns of age effects on the two
tasks are artifactual because performance on the paired
associates task is very far from ‘floor level’ and this test
has been shown to be very sensitive to deficits in several
studies on clinical patients (5, 15]. Thus,we must con-
clude that spatial working memory performance is more
sensitive to advancing age beyond 65 years than is either
paired associates learning or spatial recognition memory.
Conversely, delayed matching to sample at the longest
delay (12s) is more sensitive to age changes between 55
and 64 years,
Some trends in age-related changes in performance
were obscured by particular characteristics of CANTAB
subtests. The main example was the delayed matching to
sample test, in which there was a delay-dependent decline
in accuracy of performance from the younger age bands,
but not in older age groups (ig. 2a). Inspection of fig-
ure 2a and additional analyses suggest that performance
accuracy significantly declines as delay intervals increase
from 0 to 4 but thereafter reaches a baseline level of
about 75% correct, with the result that the delay function
is compressed. The most obvious explanation of this phe-
nomenon is that subjects are using a mnemonic strategy
which allows them to retain high levels of accuracy even at
long delays. For example it seems likely that they encoded
the patterns in terms of verbal labels derived from their
colours. This would have the effect that choice at match-
ing would be reduced from 4 to 2 stimuli; i. to a choice
between the target stimulus and the same-coloured dis-
tractor. Evidence for this is that subjects’ errors showed
more confusions between targets and same colour distrac-
tors than between targets and same shape distractors. The
superior performance of the high 1Q group 1Q3 may also
be taken to reflect their greater use of efficient verbal
encoding strategies, although it should be pointed out that
the test as a whole was relatively insensitive to differences
associated with IQ test scores. The verbal encoding hy-
pothesis is testable, since if it is correct the imposition of
an articulatory suppression task concurrently with the
DMTS task should interfere with verbal recoding and
result in sharply reduced performance.
As might be expected from many earlier demonstra-
tions that targets among distractors are located by serial
self-terminating search [43] there was a very strong linear
relationship between target detection time and the num-
ber of items on the displays in the simultaneous matching
to sample (visual search) task. Younger subjects made few
278
Rabbitt
Robbins/James/Owen/Sahakian/Melnnes/
CANTAB: Factor Analysis of @ Large
Sample errors and there was some evidence that age not only
brought about an overall slowing of performance but also
altered processing time for individual items on displays,
However, this was the opposite of the usual age com-
plexity interaction documented by Cerella [44], since the
slopes of the reaction time set size Functions reduced rath-
er than increased in the older age groups. These older
groups also showed significant increases in error rates
with set size, suggesting that as displays became more
complex they increasingly began to sacrifice accuracy to
maintain speed. Once again it appears that choice reac-
tion time tasks do not yield ‘pure’ indices of individual
differences in a single performance characteristic, ‘infor-
mation processing speed’. Latencies and accuracies must
be analysed together, as joint indices that reflect the par-
ticular compromise that individuals choose to adopt be-
tween the maintenance of speed and of accuracy. In the
matching to sample (visual search) task the oldest groups
‘were not simply slower than the young; they also adopted
different speed-error trade-off criteria. It is particularly
interesting that differences in trade-off strategy in older
groups cannot be put down to declines in their general
cognitive ability, as assessed by IQ test performance. The
low IQ group (IQ1) did not behave like elderly subjects
and there was no suggestion that the choice of speed-error
trade-off strategy varied between IQ test score groups.
Examination of gender differences showed that there
were some significant differences that could not be ex-
plained in terms of the slightly higher 1Q test score of the
male sample or in terms of age differences. While there
‘was no significant difference in accuracy between men
and women on the pattern recognition task, men were
slightly but significantly better than women at spatial rec-
ognition. Men also had a clear advantage over women on
the spatial working memory task. This cannot be ex-
plained as a consequence of a subject selection bias
favouring men, since there was no difference in overall IQ
test scores between the men and women in this popula-
tion. Further, in contrast to their poorer performance at
spatial tasks, women were slightly but significantly faster
than men at simultaneous matching to sample. Previous
studies have found that a female advantage in informa-
tion processing rate gradually emerges as age increases
within the range represented in this population. This has
been taken as evidence that the well known advantage in
longevity enjoyed by women over men entails a corre-
spondingly slower rate of central nervous system ageing,
[45]. It therefore seems possible to describe this pattern of
results as evidence for a functional dissociation between
two performance indices, spatial memory and informa-
tion processing speed, both of which have been shown in
this study to be exceptionally sensitive to biological age-
ing. It can be argued that women age more slowly than
men, and so retain their information processing ability
until later in life, but apparently continue in old age to
suffer from the slight, overall disadvantage in spatial
problem solving that has been repeatedly demonstrated
over the last 50 years [see 46 for a review]
In order to justify its use in demented populations as a
possible clinical tool, itis important not only to standard-
ize the battery on a large sample of elderly volunteers, as
swe have done here, but also demonstrate its utility in the
detection and diagnosis of dementia, in comparison with
commonly used clinical instruments, such as Folstein’s
MMSE [34]. This would clearly require a longitudinal
study which is ongoing and cannot be reported in detail at
present. However, itis of interest to compare the perfor-
mance of patients at the low (
También podría gustarte
- Disfunción Ejecutiva-Emocional en El TDAHDocumento2 páginasDisfunción Ejecutiva-Emocional en El TDAHKarina NoelAún no hay calificaciones
- Richaud 2012 - Cuestionario EmpatiaDocumento47 páginasRichaud 2012 - Cuestionario EmpatiaKarina NoelAún no hay calificaciones
- Carboni 2011 - TDAH en URUGUAYDocumento38 páginasCarboni 2011 - TDAH en URUGUAYKarina NoelAún no hay calificaciones
- Carboni 2011 - TDAH en URUGUAYDocumento38 páginasCarboni 2011 - TDAH en URUGUAYKarina NoelAún no hay calificaciones
- Control Cognitivo de La Emocion Mediante FEDocumento16 páginasControl Cognitivo de La Emocion Mediante FEKarina NoelAún no hay calificaciones
